A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 22:45 UTC
[Bugfix] Fix int32 overflow in DeepGEMM SiLU/mul FP8 Triton kernel (#42201)
Signed-off-by: vensen <vensenmu@gmail.com> Signed-off-by: Vensen <vensenmu@gmail.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Vensen committed
May 12, 2026 at 01:52 UTC
6fdb49392e5183c310c4703b883e7edbf25f6a96
1 file changed
+3
-3
vllm/model_executor/layers/quantization/utils/fp8_utils.py
+3
-3
@@ -171,7 +171,7 @@ def _silu_mul_quant_fp8_packed_kernel(
171
172
pid_pack = tl.program_id(0)
173
pid_m = tl.program_id(1)
174
- m_offset = pid_m * BLOCK_M
174
+ m_offset = pid_m.to(tl.int64) * BLOCK_M
175
176
if m_offset >= M:
177
return
@@ -321,8 +321,8 @@ def _silu_mul_per_token_group_quant_fp8_colmajor(
321
pid_n = tl.program_id(1)
322
N_2 = N // 2
323
324
- m_offset = pid_m * BLOCK_M
325
- n_offset = pid_n * BLOCK_N
324
+ m_offset = pid_m.to(tl.int64) * BLOCK_M
325
+ n_offset = pid_n.to(tl.int64) * BLOCK_N
326
if m_offset >= M:
327
return
328